Yuxuan Yao | 姚宇轩



profile photo

Biography

I am a senior undergraduate student at the School of Data Science, Fudan University. I will pursue a Ph.D. at Fudan University and Shanghai Innovation Institution, supervised by Prof. Siyu Zhu, Dr. Jingdong Wang and Prof. Yuan Qi. I previously worked with Prof. Li Zhang.

  • Outstanding Graduate of Shanghai, 2026.05
  • Fudan University College Star (10 recipients university-wide), 2026.04
  • National Scholarship, 2025
  • National Scholarship, 2023
  • Fudan University Outstanding Student Scholarship, 2024

Research Interest

  • Computer Vision
  • Generative Models

Projects & Publications

Prompt Reinjection method overview
Image Generation

Prompt Reinjection: Alleviating Prompt Forgetting in Multimodal Diffusion Transformers

Authors: Yuxuan Yao*, Yuxuan Chen*, Hui Li, Kaihui Cheng, Qipeng Guo, Yuwei Sun, Zilong Dong, Jingdong Wang, Siyu Zhu†

Venue: ICML 2026

Description: We identify prompt forgetting in multimodal diffusion transformers and introduce a training-free reinjection method that restores shallow-layer text features in deeper blocks, improving instruction following with minimal overhead.

BARD-VL benchmark comparison
Vision-Language Model

BARD: Bridging AutoRegressive and Diffusion Vision-Language Models Via Highly Efficient Progressive Block Merging and Stage-Wise Distillation

Authors: Baoyou Chen, Hanchen Xia, Peng Tu, Haojun Shi, Liwei Zhang, Yuxuan Yao, Weihao Yuan, Siyu Zhu†

Venue: ACM MM 2026 Oral

Description: BARD converts pretrained autoregressive vision-language models into efficient diffusion VLMs via progressive supervised block merging and stage-wise intra-dVLM distillation, achieving up to 3× decoding throughput while preserving multimodal performance.

3D Reconstruction
3D Reconstruction

Reflective Gaussian Splatting

Authors: Yuxuan Yao*, Zixuan Zeng*, Chun Gu, Xiatian Zhu†, Li Zhang†

Venue: ICLR 2025

Description: Ref-Gaussian is a framework for real-time high-quality rendering that integrates physically based deferred rendering and Gaussian-grounded inter-reflection to unify reflective and non-reflective scene reconstruction.

3D Reconstruction
3D Reconstruction

IRGS: Inter-Reflective Gaussian Splatting with 2D Gaussian Ray Tracing

Authors: Chun Gu, Xiaofei Wei, Zixuan Zeng, Yuxuan Yao, Li Zhang†

Venue: CVPR 2025

Description: In this paper, we introduce inter-reflective Gaussian splatting for inverse rendering. We apply the full rendering equation without simplification and compute incident radiance on the fly using the proposed differentiable 2D Gaussian ray tracing.